Duplicate QA for
Geo knowledge graphs

An open-source curator copilot for Geo (GRC-20). It scans a space, scores duplicate entities with URL-identity and relation-shape signals, and turns confirmed clusters into governance-ready merge proposals.

Pilot case study
31,800+entities scanned
1,160duplicates confirmed
561clusters across two flagship spaces
56merges in the live pilot proposal
Dedup pilot v2 — merge 56 duplicate entities
Crypto datasets · proposed from a member account
…
loading status…
Scan a space

Pick a flagship space — full nightly scan, served instantly — or paste any Geo space id. Live scans paginate the entire space (sanity cap 50,000 entities) and are queued.

How it works
  1. Snapshot — cursor-paginated GraphQL pull of the space (both Geo API dialects).
  2. Score — two-tier URL identity (report URLs vs profile links), case- and punctuation-insensitive name matching, relation-shape conflicts; transitive clustering only over ≥0.9 edges.
  3. Guard — schema entities (types and property definitions) and governance proposals are never auto-merged, only flagged for review.
  4. Publish — GRC2 binary edit → IPFS → SpaceRegistry proposal from your Geo account; editors vote, the merge applies.